Arizal, machine learning researcher focused on computer vision and image segmentation in Sragen, Central Java

.:-=+*#%@ SEGMENTATION @%#*+=-:. DEEP LEARNING .:-=+*#%@ COMPUTER VISION @%#*+=-:. PYTORCH .:-=+*#%@ DEPLOYMENT @%#*+=-:. .:-=+*#%@ SEGMENTATION @%#*+=-:. DEEP LEARNING .:-=+*#%@ COMPUTER VISION @%#*+=-:. PYTORCH .:-=+*#%@ DEPLOYMENT @%#*+=-:. .:-=+*#%@ SEGMENTATION @%#*+=-:. DEEP LEARNING .:-=+*#%@ COMPUTER VISION @%#*+=-:. PYTORCH .:-=+*#%@ DEPLOYMENT @%#*+=-:. .:-=+*#%@ SEGMENTATION @%#*+=-:. DEEP LEARNING .:-=+*#%@ COMPUTER VISION @%#*+=-:. PYTORCH .:-=+*#%@ DEPLOYMENT @%#*+=-:.

[ Hi, I'm Arizal ]

I teach models to see where one thing ends and the next begins, pixel by pixel, and turn raw images into clean, measurable masks.

Scroll down. The object beside you changes form and spins.

Role: Machine learning researcher
Specialty: Image segmentation
Base: Sragen, Central Java, UTC+7

Work

[ Selected: 5 ]

01Research / Publication2026

[ ICORIS 2026 · Scopus-indexed · First author ]

Four Lenses, One Verdict: AI-Image Detection

Paper at ICORIS 2026 (DOI pending). A CNN trained from scratch that fuses RGB, SRM noise, FFT and Sobel edges with attention to spot AI-generated images: 96.9% on CIFAKE, no pretrained weights.

Read the writeup ↗

02Kaggle / AI agent security2026

[ Silver medal · Kaggle notebook ]

Multi-Turn Exfiltration with Adaptive Burst

Silver medal notebook in the OpenAI Kaggle competition on AI Agent Security (Multi-Step Tool Attacks), studying multi-turn data exfiltration with an adaptive burst approach.

View on Kaggle ↗

03Kaggle / Classification2026

[ Bronze medal · Kaggle notebook ]

Cracking the Generator (0.944 AUC)

Bronze medal notebook in Kaggle Playground Series Season 6, Episode 9. Reaches 0.944 AUC by working out how the data generator behaves.

View on Kaggle ↗

04Kaggle / Classification2025

[ Bronze medal · Kaggle notebook ]

Kaggle S5E7: Bronze Medal

Bronze medal in Kaggle Playground Series S5E7. Ensemble modelling and feature engineering for a binary classification task.

View on Kaggle ↗

05Computer vision / Detection2025

Neural Traffic Analytics System

Real-time traffic monitoring that detects and tracks vehicles and pedestrians with DETR and the SORT algorithm.

View project ↗

[ Archive ]All projects12 entries

Profile

[ who is behind the model ]

I'm Arizal, a machine learning researcher focused on computer vision and image segmentation, based in Sragen, Central Java. I care about the boundary between one object and the next, and about models that get it right on messy real-world data.

I enjoy the whole loop: collecting and labelling data, choosing and training the architecture, measuring what the numbers do not show, and shipping a model that someone can actually use. Outside of projects I read papers, join competitions and build small tools for visualising model output.

Image segmentationSemantic, instance and panoptic segmentation with PyTorch, from U-Net style baselines to transformer based models.
Training and evaluationLoss design, augmentation, class imbalance and honest metrics such as Dice, IoU and boundary error.
Data and annotationLabelling guidelines, tooling and active learning loops that cut the cost of every new mask.
Deployment and optimisationExport to ONNX, quantisation and distillation, so models run on servers and on small devices.

Experience

2026
Speaker, machine learning workshopSpeaker · Campus eventInvited to speak and present at a machine learning workshop organised by a student technology organisation at my campus.
2026
Machine Learning Engineer Intern, VINIX 7Internship · Remote · 4 monthsMBKM Batch 5 independent study and internship, placed in the Machine Learning Engineer division.
2026
Teaching assistant, STMIK Amikom SurakartaPart-timeAssistant for the course Data Preprocessing for AI.
2025
Software developer, Jasmin Motor SoloSoftware projectBuilt software that saves receipts created on the computer straight into an Excel file or recap file.
2024
Research assistant, lecturer's research projectResearch · Web app · Semester 1Built a web app where students send complaints and suggestions about lecturers. The best-rated lecturers are ranked with a decision support system (DSS), and lecturers get improvement suggestions from an LLM (via the Groq API) that reads the student feedback.
2023–2024
Data entry, Jasmin Motor SoloPart-time · About 1 yearFirst job after vocational high school (SMK). Typed data from paper receipts into Excel, visiting the office once a month.

Like a segmentation mask, every cell on this page gets exactly one label: one of ten characters, ordered from dark to light. The denser the character, the brighter the dot.